Vagal Impairment and Cardiovascular Risk in Those with Zero to Low Coronary Artery Calcification Scores: The Multi-Ethnic Study of Atherosclerosis

Abstract

Coronary artery calcification (CAC) is the most reliable noninvasive predictor of major adverse cardiovascular events (MACE). Individuals with no detectable or minimal CAC (Agatston score 1–100) are considered at the lowest risk. However, MACE do occur in these groups. This study evaluated whether quantification of cardiac vagal activity by heart rate fragmentation (HRF) improved MACE risk prediction beyond CAC imaging. Our study population is a cohort of the Multi-Ethnic Study of Atherosclerosis (MESA). Cox regression models were used to assess the association between HRF, derived from polysomnographic ECGs, and incident MACE in the overall cohort with concurrent polysomnographic ECG and CAC data, and three non-overlapping subgroups: “very-low-risk” (CAC = 0), “low-risk” (0 < CAC < 100), and “higher-risk” (CAC ≥ 100). Over a median (1^st; 3^rd quartiles) follow-up period of 8.9 (8.4; 9.4) years, there were 164, 29, 47 and 88 incident MACE in the overall cohort (n=1,354), very-low-risk (n=495), low-risk (n=422) and higher-risk (n=437) subgroups, respectively. A one-standard deviation increment in HRF was associated with a 22% (3% – 44%) and a 60% (16% – 122%) increase in the rate of MACE in the overall cohort and those without detectable CAC, respectively. Neither the Framingham nor MESA-CAC index was predictive of MACE in the lowest-risk subgroup until nearly a decade of follow-up. From a physiologic perspective, our results suggest that vagal dysfunction precedes the onset of overt cardiovascular disease (CVD). From a translational perspective, they indicate that HRF enhances risk stratification, especially in populations traditionally classified as very-low risk.

NEW & NOTEWORTHY

This study is the first to show that cardiac parasympathetic function, assessed via heart rate fragmentation (HRF), is independently associated with major adverse cardiovascular events (MACE) in individuals traditionally considered low-risk due to non-detectable or minimal coronary artery calcification (CAC). HRF outperforms both the Framingham and MESA-CAC risk scores in the lowest-risk subgroup over the short-term (<10 years).

INTRODUCTION

Coronary artery calcification (CAC) is recognized as the single most reliable predictor of major adverse cardiovascular events (MACE) [1]. The Agatston score, derived from computed tomography (CT) imaging, is a measure of overall CAC burden. Individuals with zero or low (1–100) Agatston scores are often deemed at the lowest risk of MACE [2]. However, cardiovascular (CV) morbidity and mortality do occur in these groups [3, 4]. Traditional CV risk predictors, such as the Framingham Risk Score, exhibit their highest predictive accuracy in individuals at high risk, but their performance is more limited in those at low and very low risk. Enhancing risk stratification in these lower-risk populations through noninvasive, cost-effective biomarkers suitable for longitudinal monitoring remains a key challenge in contemporary cardiology and precision medicine [5].

Zero CAC score is often interpreted as evidence of robust vascular health. However, a zero CAC score reflects only the absence of identifiable calcified plaque by current imaging modalities and does not preclude functional abnormalities or non-calcified atherosclerosis. Other factors, including autonomic imbalance, particularly reduced parasympathetic (vagal) activity, may contribute to sub-clinical CV risk among those with zero CAC.

Physiologic cardiac vagal tone exerts homeostatic control over heart rate, modulates baroreflex sensitivity, and is thought to activate the cholinergic anti-inflammatory pathway, thereby buffering against sympathetic overactivation, oxidative stress, and endothelial dysfunction [6–8]. These mechanisms are implicated in the early development of atherosclerosis, even before structural changes become apparent on advanced imaging studies [9–12]. Moreover, reduced vagal activity has been associated with pro-inflammatory signaling [13, 14], diminished vascular compliance [15] and impaired cardiac electrophysiologic stability [16, 17], all of which may contribute to subclinical CV vulnerability. In this context, physiological markers of parasympathetic dysfunction may improve risk stratification in individuals without detectable anatomical changes on CT imaging studies. Heart rate fragmentation (HRF) [18, 19], which reflects non-respiratory beat-to-beat fluctuations in sinus rhythm, is emerging as a novel, noninvasive biomarker of vagal control [20–28].

The specific goal of this study was to determine whether HRF, derived from polysomnographic ECG recordings [18], added value to CAC imaging in assessing risk of MACE over relatively short follow-up periods (i.e., less than a decade) in an overall cohort of participants in the Multi-Ethnic Study of Atherosclerosis (MESA) and two subgroups classified as “very-low-risk” (CAC = 0) and “low-risk” (0 < CAC < 100). This approach was motivated by the hypothesis that vagal functionality plays an important role in the pathogenesis of atherosclerosis, particularly in regulating immune function and mitigating the deleterious effects of oxidative stress and endothelial dysfunction induced by excessive sympathetic activation [29].

METHODS

Study Population

The study population is a subset of MESA participants. The MESA is a community-based cohort study of clinical and subclinical cardiovascular disease (CVD) [30]. Briefly, over approximately two years, starting in July 2000, 6,814 persons between the ages of 45 and 84 without evident clinical CVD were recruited at six field centers in the US. Institutional review boards from each study site approved the conduct of this study, and written informed consent was obtained from all participants. There have been six subsequent MESA exams. For this investigation, we selected participants with a polysomnographic (PSG) study conducted between 2010 and 2013 in conjunction with the fifth MESA examination. Out of the 2,057 participants who completed the PSG, we excluded those with one or more of the following: poor ECG quality (n = 33), pacemaker (n = 14), atrial fibrillation at the time of the PSG (n = 22), < 2 h of combined sleep periods scored as rapid eye movement, stage 1, 2, 3, or 4 (n = 16), < 75% normal sinus beats between sleep onset and termination (n = 11), history of any CV event before the PSG (n = 185), no follow-up after the PSG study (n = 7), missing values for any of the variables need for the calculation of the Framingham index (n = 41) and those without Agatston scores (n = 450) obtained at MESA Exam 5 (2010–2012). Of the remaining 1,354 participants, 495, 422 and 437 had CAC = 0, 0 < CAC < 100 and CAC ≥ 100, respectively.

Exposure

Heart rate fragmentation was computed from the ECG channel of the PSG studies, as described in detail in [18]. The ECG channel was sampled at 256 Hz, providing a temporal resolution of 4 ms. Briefly, the time series of cardiac interbeat [normal-to-normal (NN)] intervals were mapped to a sequence of three symbols, “−1”, “1,” and “0”, representing heart rate acceleration, deceleration and no change in heart rate, respectively. The rules for the mapping were: “−1” if ΔNN < −4 ms, i.e., if an NN interval is shorter than its predecessor (heart rate acceleration); “1” if ΔNN > 4 ms, i.e., if an NN interval is longer than its predecessor (heart rate deceleration); and “0” if −4 < ΔNN < 4 ms, i.e., if the values of an NN interval and of its predecessor are the same (i.e., heart rate did not change).

The transitions between different symbols (i.e., “−11,” “1–1,” “−10,” “10,” “0–1” and “01”) are termed “inflection points.” These inflection points reflect transitions from: acceleration to deceleration, deceleration to acceleration, acceleration to no change, deceleration to no change, no change to acceleration and no change to deceleration, respectively. The percentage of inflection points (PIP), which quantifies the overall level of HRF [18], was the selected HRF measure for this study.

Clinical Follow-Up, Event Classification and CT Image Acquisition

In addition to clinical exams, participants are followed every 9–12 months to inquire about hospital admissions, CV outpatient diagnoses and procedures, and deaths. Discharge diagnosis codes are obtained for all hospitalizations and medical records are obtained when heart failure, myocardial infarction, stroke, or death are reported. Trained personnel obtain any hospital records suggesting possible CV events, which are then adjudicated by physicians. Nonfatal endpoints in MESA include congestive heart failure, angina, myocardial infarction, percutaneous coronary intervention, coronary bypass grafting or other revascularization procedure, resuscitated cardiac arrest, peripheral arterial disease, stroke (non-hemorrhagic) and transient ischemia attack. Cardiovascular deaths, as adjudicated by committee review, included fatalities directly related to stroke or coronary heart disease. For other CV deaths, the underlying cause are obtained through state or vital statistics departments. The definition and adjudication of these events have been described in detail previously [30–32]. The cut-off date for the surveillance period was December 31, 2020. The participants were scanned using either electron-beam or multi-detector CT scanners as previously described [33]. The images were transferred and read at the MESA central CT reading center.

Statistical Analysis

Continuous variables are summarized as median, first and third quartiles, unless otherwise indicated. Categorical variables are presented as numbers and percentages.

The association between independent variables and incident MACE was assessed using Cox proportional hazard analysis. Efron’s method was used to handle ties. Survival analyses were performed for follow-up periods of 5, 6, 7, 8 and 9 years. The failure time for participants who experienced a MACE during a specific follow-up period was the number of days between the date of the polysomnographic (PSG) study and the date of MACE diagnosis. For those without a MACE event, failure time was the duration of the follow-up period (e.g., 1,825 days for the 5-year follow-up analyses), or the number of days from the PSG to either death or loss to follow-up, whichever occurred first. Statistical significance was set at a p-value < 0.05.

We defined three risk groups based on CAC burden: CAC = 0, 0 < CAC < 100 and CAC ≥ 100, reflecting “very-low-risk,” “low-risk” and “higher” CV risk, respectively. Diabetes was defined as fasting glucose ≥ 126 mg/dL or the use of oral hypoglycemic medications or insulin. Hypertension was defined as systolic blood pressure ≥ 140 mmHg or diastolic blood pressure ≥ 90 mmHg or use of anti-hypertension medication. Current smoking status and medication use were determined by self-report. The estimated glomerular filtration rate, eGFR, was calculated using the 2021 Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation, which incorporates the serum levels of both cystatin C and creatinine. The apnea-hypopnea index (AHI) was calculated considering apneas and hypopneas with ≥ 3% desaturation or arousal.

The Framingham Heart Study 10-year risk index [34] is a composite index that incorporates age, sex, diabetes, total cholesterol, HDL, systolic blood pressure and current smoking. The MESA 10-year risk index of coronary heart disease with CAC (MESA-CAC) [35] includes these variables as well as lipid-lowering medication, family history of myocardial infarction and the Agatston score. Due to their skewed distributions, the risk indices, the Agatston scores and AHI were log-transformed. For the latter two, we specifically used ln (x + 1) due to the occurrence of zero values.

Both unadjusted and adjusted analyses were performed. The models adjusted for individual risk factors included age, sex, race/ethnicity, Agatston score, diabetes status, systolic blood pressure, anti-hypertensive medication, total cholesterol, HDL, lipid-lowering medication, smoking status, eGFR and AHI.

Standardized hazard ratios per one-standard deviation (SD) increase in the independent variable were calculated with associated 95% confidence intervals (CI). The assumption of proportional hazards was tested using a global test based on Schoenfeld residuals [36]. No violations were found. The predictive power of the survival models was assessed using Harrell’s C index. The likelihood ratio test was used to compare the fit of two nested models (null model vs. null model + HRF). The null hypothesis for each likelihood ratio test was that the two nested models fitted the data equally well. Rejection of the null hypothesis implied that the larger model fitted the data better, indicating that HRF added value to the null model.

To account for potential nonlinearity in the relationship between HRF and the hazard function for MACE, we repeated the analysis using restricted cubic splines with three degrees of freedom. We conducted a joint Wald test with the null hypothesis that the coefficients of the second and third spline terms were both equal to zero. Additionally, we performed a likelihood ratio test to assess whether the inclusion of the spline terms significantly improved model fit. Since neither test provided evidence of nonlinearity, we report the results from the linear models.

To quantify how CAC burden modified the association between HRF and incident MACE, we included an interaction term between HRF and CAC group in the Cox models. Additionally, we conducted stratified analyses to directly assess the association between HRF and incident MACE within each CAC subgroup. Given the relatively small number of incident MACE within the CAC subgroups, these analyses were adjusted for either the Framingham or MESA-CAC risk index rather than individual risk factors.

RESULTS

During a median follow-up period of 8.9 years (1st – 3rd quartiles: 8.4 – 9.4), a total of 164 MACE occurred. In the very-low-risk subgroup (CAC = 0, n = 495), there were 29 events (5.8%). In the low-risk (0 < CAC < 100, n = 422), and higher-risk (n = 437) subgroups, there were 47 (11.1%) and 88 (19.9%) MACE, respectively. The demographic and clinical characteristics of each study group at the fifth MESA examination are presented in Table 1. Age, HRF, systolic blood pressure, AHI, and both the Framingham and MESA-CAC 10-yr risk indices increased progressively across CAC groups, from the group with zero CAC to the one with CAC ≥ 100. The proportion of males, participants with diabetes and those using antihypertensive or lipid-lowering medications also increased across CAC groups. In contrast, HDL and eGFR decreased across CAC groups. Overall, participants without detectable CAC exhibited a healthier risk profile.

Table 1.

Demographic, clinical and other characteristics of study groups.

	Full cohort	CAC = 0	0 < CAC < 100	CAC ≥ 100
N	1,354	495	422	437
HRF (%)	58.0 (53.3 – 62.9); 58.2 ± 7.0	56.5 (51.9 – 61.3); 56.8 ± 6.8	57.4 (53.1 – 62.7); 58.0 ± 6.9	60.1 (55.5 – 64.8); 60.0 ± 7.1
Age (yrs)	67 (60 – 75); 67.8 ± 9.0	62 (58 – 69); 63.9 ± 7.5	66 (61 – 75); 67.8 ± 8.84	72 (65 – 78); 72.2 ± 8.7
Male	620 (45.4)	153 (30.6)	211 (49.8)	256 (57.8)
Race White	492 (36.0)	156 (31.2)	149 (35.1)	604 (46.3)
Chinese	173 (12.7)	71 (14.2)	41 (9.7)	61 (13.8)
Black	374 (27.4)	157 (31.4)	122 (28.8)	95 (21.4)
Hispanic	328 (24.0)	116 (23.2)	112 (26.4)	100 (22.6)
Agatston score*, ln	20.2 (0 – 163.4); 2.91 ± 2.57		27.6 (11.0 – 57.3); 3.19 ± 1.01	348 (171 – 687); 5.92 ± 0.88
Antihypertensive Rx	678 (49.6)	176 (35.2)	224 (52.8)	278 (62.8)
Systolic BP (mmHg)	120 (108 – 134); 122.5 ± 20.3	117 (105 – 131); 119.6 ± 19.9	119 (110 – 133); 122 ± 19.2	123 (111 – 139); 126 ± 21.2
Diabetes	244 (17.9)	65 (13.1)	61 (15.8)	112 (25.4)
Total cholesterol (mg/dL)	186 (162 – 208); 186.5 ± 35.4	191 (168 – 214); 192.7 ± 35.3	186 (162 – 210); 187.5 ± 34.5	177 (155 – 201); 178.4 ± 34.8
HDL (mg/dL)	53 (45 – 64); 55.9 ± 16.2	55 (46 – 67); 57.8 ± 16.3	53 (44 – 62); 55.2 ± 17.0	52 (43 – 63); 54.3 ± 15.1
Lipid-lowering Rx	461 (33.7)	102 (20.4)	149 (35.1)	210 (47.4)
Smoking (current)	34 (6.8)	34 (6.8)	24 (5.7)	29 (6.6)
AHI* (events/hr), ln	17.5 (8.77 – 32.3); 2.85 ± 0.88	14.5 (6.78 – 27.4); 2.65 ± 0.90	18.5 (9.53 – 33.0); 2.89 ± 0.89	20.9 (12.2 – 36.8); 3.04 ± 0.79
eGFR (mL/min/1.73 m2)	91.4 (77.2 – 103); 89.0 ± 18.8	94.2 (82.2 – 104); 92.1 ± 16.1	93.0 (77.5 – 104); 89.5 ± 19.6	86.7 (71.6 – 100); 84.9 ± 20.0
Framingham risk* (%), ln	13.3 (7.72 – 22.9); 2.55 ± 0.77	9.29 (5.13 – 15.5); 2.19 ± 0.76	13.7 (8.56 – 22.1); 2.60 ± 0.72	19.6 (11.7 – 30.7); 2.91 ± 0.65
MESA-CAC risk * (%), ln	5.25 (2.24 – 11.2); 1.63 ± 0.98	1.91 (1.34 – 2.62); 0.63 ± 0.47	5.99 (4.06 – 8.02); 1.75 ± 0.55	14.0 (9.79 – 19.3); 2.64 ± 0.49
Incident MACE – yr 5	93 (6.8)	14 (2.8)	28 (6.6)	51 (11.5)
Incident MACE – yr 6	112 (8.2)	17 (3.4)	31 (7.3)	64 (14.4)
Incident MACE – yr 7	128 (9.4)	20 (4.0)	36 (8.5)	72 (16.3)
Incident MACE – yr 8	149 (10.9)	24 (4.8)	44 (10.4)	81 (18.3)
Incident MACE – yr 9	164 (12.0)	29 (5.8)	47 (11.1)	88 (19.9)

^*For values shown are the median (1st – 3rd quartiles) of the original (untransformed) variable and the mean ± standard deviation of the natural logarithmic-transformed variable. Abbreviations: AHI, apnea-hypopnea index; CAC, coronary calcification score; eGFR, estimated glomerular filtration rate; HDL, high density lipoprotein; HRF, heart rate fragmentation; MACE, major adverse cardiovascular event; MESA-CAC, MESA 10-year risk index of coronary heart disease with CAC; Rx, medication.

Increased HRF was independently associated with a higher rate of MACE in the overall cohort. In a Cox proportional hazards model adjusted for individual risk factors, a one-SD increment in HRF was associated with a 22% (95% CI: 3% – 44%) higher rate of MACE over a nine-year follow-up period. Notably, the inclusion of HRF significantly improved the performance of the model with only the risk factors. The C-index increased by nearly 1% (∆C = 0.008, p = 0.022). The inclusion of HRF in models with the Framingham or MESA-CAC also improved their performance, with the C-index increasing by 0.012 (p = 0.004) and 0.008 (p = 0.020), respectively.

To assess the extent to which CAC burden modified the association between HRF and incident MACE, we included an interaction term between HRF and CAC group in the Cox proportional hazards model adjusted for individual risk factors. The results indicated that the relationship varied across CAC groups, with the strongest association observed in participants without detectable CAC (Table 2). In this subgroup, a one-SD increment in HRF was associated with a 60% (95% CI: 16% – 122%) higher rate of MACE over a nine-year follow-up period. In the low-risk subgroup (0 < CAC < 100) the association did not reach statistical significance for longer (≥ 8 years) follow-up periods. In the higher risk subgroup (CAC ≥ 100), HRF was not independently associated with incident MACE regardless of follow-up duration.

Table 2.

Hazard ratios (HR) for HRF derived from multivariable models Cox regression models with an interaction term between HRF and CAC group. The covariates in the models were age, sex, race/ethnicity, Agatston score, diabetes status, systolic blood pressure, anti-hypertensive medication, total cholesterol, HDL, lipid-lowering medication, smoking status, eGFR and the AHI index.

CAC burden group	Follow-up (years)	HR	95% CI		P value
CAC = 0	5	1.88	1.23	2.88	0.004
	6	1.84	1.24	2.72	0.002
	7	1.71	1.18	2.47	0.004
	8	1.64	1.16	2.33	0.006
	9	1.60	1.16	2.22	0.005
0 < CAC < 100	5	1.54	1.08	2.20	0.017
	6	1.49	1.05	2.10	0.024
	7	1.41	1.02	1.94	0.037
	8	1.32	0.98	1.77	0.070
	9	1.31	0.98	1.74	0.070
CAC ≥ 100	5	1.10	0.82	1.48	0.520
	6	1.12	0.86	1.47	0.388
	7	1.06	0.83	1.36	0.650
	8	1.04	0.82	1.31	0.749
	9	1.04	0.83	1.30	0.738

Abbreviations: AHI, apnea-hypopnea index; CAC, coronary calcification score; CI, confidence interval; eGFR, estimated glomerular filtration rate; HDL, high density lipoprotein; HR, hazard ratio; HRF, heart rate fragmentation.

The results of stratified analyses by CAC group are presented in Figure 1. In the very-low-risk subgroup, the Framingham risk index was not significantly association with incident MACE, and the MESA-CAC risk index was only associated when the follow-up period extended to nine years (B, F and J). In contrast, HRF exhibited the strongest association with incident MACE in this subgroup (B, F, and J). The significance of the association persisted after adjusting for the MESA-CAC risk index, including for the longest follow-up period (J).

Figure 1.

Cox proportional hazards regression models of MACE employing the cohort of MESA participants with polysomnographic ECGs and CAC assessments obtained in conjunction with Exam 5 (2020 – 2012) (A, E & I), and the subgroups of those with very-low (CAC = 0) (B, F & J), low (0 < CAC < 100) (C, G & K) and higher (CAC ≥ 100) (D, H & L) CV risk. The top panels (A-D) show the results from univariable models. The middle panels (E-H) show the results of models with HRF and the Framingham risk index. The bottom panels (I-L) show the results of models with HRF and the MESA-CAC risk index. The symbols and error bars indicate the hazard ratios and 95% CI, respectively.

As shown in Figure 2, the rate of MACE increased monotonically with the degree of HRF. The association between HRF and MACE was significant both above and below the reference point (the mean HRF value), in the overall cohort (A and D) and the very-low-risk subgroup (B and E). In the low-risk subgroup, the association was only borderline significant (C and F). In the higher-risk subgroup, while still positive (higher HRF, higher risk of MACE), the association was not significant (not shown).

Figure 2.

Hazard ratios for the risk of MACE as a function of HRF derived from Cox proportional hazards regression models. The analyses were performed for the overall cohort of MESA participants with polysomnographic ECGs and CAC assessments (A and D), and the subgroups of those with very-low (CAC = 0) (B and E) and low (0 < CAC < 100) CV risk (C and F). The follow-up time was nine years. The top panels (A-C) show the results of unadjusted analyses. The bottom panels (D-F) show the results of the analyses adjusted for the MESA-CAC risk index. The blue lines and shaded areas indicate the hazard ratios and the 95% confidence intervals, respectively. The horizontal dotted lines mark the hazard ratio of 1.0. The histograms shown on the top panels (A-C) display the distributions of HRF values within each study groups. The y-axis range (not shown) is the same (0–90 counts) for all the histograms.

DISCUSSION

This MESA study demonstrates the value of HRF, a marker of reduced vagal functionality, as a midterm independent predictor of incident MACE. We analyzed the cohort of MESA participants with CAC assessment and polysomnographic ECG recordings obtained in conjunction with Exam 5 (2010–12) and very-low, low and higher CV risk subgroups, defined by the absence of detectable CAC, the presence of a small amount (0 < CAC < 100) and of a larger amount (CAC ≥ 100), respectively.

Our two key findings were that increased HRF was independently associated with a higher rate of MACE in the overall cohort and that this association was strongest in the subgroup of participants without detectable CAC. Notably, this subgroup is the one in which HRF may be most valuable, as neither the Framingham nor the MESA-CAC risk index predicted MACE until nearly a decade of follow-up.

We previously showed that HRF was significantly associated with incident MACE in a MESA study with a follow-up period of 3.0 ± 0.6 years during which 72 (4.0%) MACE occurred [37]. Traditional heart rate variability (HRV) metrics were not associated with incident MACE. These findings were independently confirmed by other researchers in the Swiss HypnoLaus population-based study [22], which had a comparable follow-up period (4.1 ± 1.1 years) and number of adverse events [68 (3.8%)]. Since our 2018 publication [37], the follow-up time and number of incident events in MESA have both tripled, enabling survival analyses in low-risk subgroups and allowing for a more comprehensive set of adjustments in analyses with the overall cohort.

Mechanisms underlying the association between vagal dysfunction and incident MACE

The finding that the association between HRF and MACE rate remained significant even after adjusting for major risk factors such as age, diabetes, hypertension and CAC suggests that autonomic dysfunction may play an independent and mechanistically distinct role in CV pathophysiology.

The efferent vagus nerve modulates systemic inflammation through the cholinergic anti-inflammatory pathway (CAP) [6,7], which operates via a neuroimmune reflex referred to as “vago-splenic signaling” [38–43]. In this pathway, vagal efferent signals are relayed through the celiac ganglion to stimulate splenic sympathetic nerves, which release norepinephrine onto β2-adrenergic receptors on resident T cells. These T cells, in turn, produce acetylcholine (ACh), which binds to α7 nicotinic ACh receptors (α7nAChR) on macrophages and other innate immune cells, suppressing the release of proinflammatory cytokines such as TNF-α, IL-6, and IL-1β. This vagally-mediated cascade is thought to play a critical role in limiting systemic inflammation and may protect against inflammation-driven myocardial injury and remodeling [41–43].

Complementing this systemic pathway, emerging evidence supports a cardioprotective role for the non-neuronal cardiac cholinergic system (NNCCS) [44, 45]. This intrinsic signaling network comprises resident cardiac cells, including cardiomyocytes and fibroblasts, which are capable of producing and responding to ACh. Notably, the NNCCS can be activated by physiological stressors such as ischemic preconditioning, suggesting that it may function as an endogenous mechanism of myocardial protection [44]. Although not directly innervated by vagal fibers, the NNCCS may be functionally influenced by parasympathetic tone, raising the possibility that systemic parasympathetic activity and local cholinergic signaling act in concert to regulate myocardial responses to stress and injury.

Autonomic dysfunction may also contribute to CV pathology through other interrelated mechanisms. The autonomic nervous system operates as an exquisitely regulated network in which increased efferent activity in one branch is typically accompanied by a reciprocal decrease in the other one – a dynamic “see-saw” relationship [46, 47]. Reduced vagal tone may therefore fail to counterbalance sympathetic overactivity, which has been implicated in proinflammatory signaling, oxidative stress, and endothelial dysfunction [10, 48]. Moreover, vagal functionality may influence CV risk through its interactions with the hypothalamic-pituitary-adrenal axis, modulating stress responses and preventing dysregulated cortisol production, a factor that exacerbates vascular inflammation and endothelial impairment when excessive [49]. Vagal tone is also a critical determinant of baroreflex sensitivity, which helps regulate short-term blood pressure fluctuations [50], and of electrophysiological stability [17]. These mechanisms highlight the therapeutic potential of targeting parasympathetic pathways in CV disease.

Further support for links between parasympathetic-immunoregulatory interactions comes from experimental models, in which both electrical vagus nerve stimulation and pharmacological activation of α7nAChR have demonstrated anti-inflammatory effects in a wide range of conditions (e.g., atherosclerosis, inflammatory bowel disease and rheumatoid arthritis) [8, 13, 40, 51–53].

Composite risk indices versus dynamical variables

Traditional CV risk models, such as the Framingham Risk Score and MESA-CAC, perform well in intermediate-risk individuals but are less reliable in those with minimal or no CAC. While traditional risk factors are either binary (e.g., diabetes status) or snapshot measures of continuous variables (e.g., office visit systolic blood pressure), HRF is a continuous variable that quantifies a dynamical process (short-term autonomic regulation of heart rate). HRF has two additional desirable characteristics for risk stratification: a bell-shaped distribution (Fig. 2, panels A–C) and a monotonic relationship with MACE rate (Fig. 2, panels A–F). In the analyses presented here, the association between HRF and incident MACE was significant for any level of HRF. This finding suggests that both low and high HRF levels carry prognostic information, the former as a negative risk factor for MACE and the latter as a positive one.

Unlike CAC, which reflects cumulative atherosclerotic burden and changes slowly over time, HRF, as a dynamical index of integrative autonomic functionality, can be more responsive to longitudinal changes in autonomic status due to lifestyle modifications, medications, or disease progression. Given these properties, HRF has potential translational utility for early identification of at-risk individuals, longitudinal monitoring of CV risk, evaluation of the efficacy of therapeutic interventions, and guidance of early strategies in preventive cardiology.

Heart rate fragmentation versus heart rate variability

The parasympathetic branch of the autonomic nervous system is responsible for the coupling between heart rate and respiration, which produces an oscillatory pattern in heart rate termed “respiratory sinus arrhythmia,” (RSA). This pattern is characterized by a graduated increase in heart rate during inspiration and decrease during expiration. Recently, the term respiratory heart rate variability (RespHRV) has been proposed to replace RSA, avoiding the connotation of an abnormality [54]. For decades, investigators have used metrics of amplitude of variability, such as the root mean square of successive differences (RMSSD) and high-frequency (HF) power of cardiac interbeat interval time series to quantify vagal tone modulation. However, with aging and disease, fluctuations in heart rate may emerge at frequencies above the respiratory range [18, 53]. Regardless of their amplitude, these fluctuations are a manifestation of the breakdown of cardiopulmonary coupling and a dynamical marker of autonomic dysfunction.

The existence of two distinct types of variability, one physiologic (RespHRV) and the other pathologic (HRF) is of importance because it may lead to misinterpretation of traditional HRV analysis. Indeed, from the values of RMSSD and HF power one cannot infer whether variability stems from physiologic vagal modulation or pathologic cardiopulmonary decoupling, i.e., fragmentation. This limitation renders traditional HRV metrics inherently ambiguous, particularly for analyses of middle-aged and older individuals, who are at the highest risk for MACE. We developed a conceptual framework and methodology for the quantification of these ultra-high-frequency fluctuations under the rubric of heart rate fragmentation, HRF [18, 29, 37, 55].

Notably, HRF is robust to outliers since it is not influenced by the magnitude of individual data values but rather by the temporal structure of variability. For example, the two alternating sequences “800, 910, 790, 820” and “1000, 1200, 1000, 1100,” with different means and SDs have the same value of HRF given that their temporal structure is the same: an increase, followed by a decrease, followed by an increase in value. In contrast, even a small percentage of outliers may substantially change HRV metrics.

Potential translational applications of HRF

Our finding that HRF was associated with incident MACE in participants with zero or low CAC scores, independently of both modifiable and non-modifiable risk factors, suggests that HRF screening could help identify individuals who might benefit from early preventive strategies, such as physical activity, weight loss, vagus nerve stimulation and mindfulness-based interventions. Notably, as a dynamical marker extractable from wearable technology, HRF could help personalize such interventions and provide a means to monitor their effectiveness over time.

LIMITATIONS AND FUTURE STUDIES

HRF, designed to overcome key limitations of traditional HRV measures, namely their susceptibility to confounding by non-vagal, ultra-fast fluctuations in sinus rhythm, remains a relatively novel metric. However, the consistent association of HRF with MACE in two independent studies (MESA and HypnoLaus [22]) provides compelling support for its relevance as a biomarker of subclinical CV vulnerability. Broader validation across diverse populations, age groups, and measurement settings will be necessary to determine its generalizability and clinical utility. We also note that the subgroup analyses by CAC level were limited by relatively modest event counts, especially in the very-low-risk group (CAC = 0). As such, effect estimates within subgroups should be interpreted with caution. Future studies are needed to examine the prognostic value of HRF assessed from diurnal and shorter ECG recordings, as compared to the overnight measures reported here. Additionally, investigating whether longitudinal changes in HRF provide incremental prognostic information will be of interest. Finally, the observational design of MESA does not permit causal inference regarding the direct role of vagal impairment in the pathogenesis of MACE.

CONCLUSIONS

This MESA study shows that impaired vagal function, non-invasively assessed by HRF, is independently associated with incident MACE over intermediate follow-up periods. Notably, HRF showed the strongest predictive value in individuals without detectable CAC, who are typically considered at lowest CV risk. The findings suggest that HRF reflect latent pathophysiological processes that contribute both to the development of atherosclerosis and to heightened short- and intermediate-term CV risk, even in the absence of calcified plaque. By quantifying pathological fluctuations in autonomic regulation not reflected in standard HRV metrics or anatomical imaging, HRF also offers a complementary dimension to risk stratification. These findings support further investigation into HRF’s prognostic value across settings and its responsiveness to interventions aimed at preserving or restoring autonomic function.

Non-standard Abbreviations and Acronyms

α7nAChR

α7 nicotinic acetylcholine receptors

Ach

acetylcholine (Ach)

AHI

apnea-hypopnea index

CAC

coronary calcification

CAP

cholinergic anti-inflammatory pathway

CI

confidence interval

CV

cardiovascular

CVD

cardiovascular disease

eGFR

estimated glomerular filtration rate

HF

high-frequency

HRF

heart rate fragmentation

MACE

major adverse cardiovascular events

MESA

Multi-Ethnic Study of Atherosclerosis

PSG

polysomnographic study

RMSSD

root mean square of successive differences

SD

standard deviation

Data Sharing Statement

Data will be made available upon approval by the MESA Publications and Presentations Committee of a manuscript proposal (https://www.mesa-nhlbi.org, accessed February 1, 2025).